Vowel Formant Frequencies Research Articles

Acoustic and intelligibility characteristics of sentence production in neurogenic speech disorders

Although sentence production is more common than single-word production in daily life, acoustic and intelligibility data from persons with neurogenic speech disorders is largely limited to single-word and target-word utterances. The purpose of the current study is to examine sentence productions of individuals with Parkinson’s disease (PD) and amyotrophic lateral sclerosis (ALS), and to compare them to data obtained from neurologically normal individuals. Each speaker produced multiple repetitions of a small set of sentences. Acoustic measures included overall sentence durations, segment durations, and vowel formant frequencies. Speech intelligibility of the sentences was measured using scaling techniques. For some subjects, previously obtained acoustic and intelligibility data for single words will be related to the findings for sentences. Results will be discussed descriptively as deviations from patterns observed in normal individuals, and in terms of associations between the acoustic and intelligibility data. [Work supported by NIDCD Award DC00319.]

The effect of speaking-rate manipulations on segment durations and formant frequencies in dysarthric speakers with amyotrophic lateral sclerosis

The effect of speaking-rate manipulations on dysarthric speech is interesting from both theoretical and clinical perspectives. The purpose of the present study was to examine the influence of speaking-rate adjustments (conversational versus fast) on segment durations and vowel formant frequencies derived from sentence productions in speakers with amyotrophic lateral sclerosis (ALS), a disease that involves death of motor neurons serving the speech mechanism, with subsequent loss of muscle mass and strength. Speakers with ALS were clearly able to change their rate for the fast condition, though not in the same proportion as the control speakers. The segment durations responded to the rate manipulation in much the same way. For formant frequencies, the effect of rate was complex, but may be generally summarized as follows: F1 was affected systematically by the rate manipulation in control males, but not among the male subjects with ALS; for females in both groups, there was little effect of rate on F1. The effects of rate on F2 measures showed no clear trend, varying across vowels and groups. Results will be discussed relative to the speech production deficit in ALS and acoustic models of speech intelligibility. [Work supported by NIDCD Award DC00319.]

Perceptual evaluation of alternate formant normalization models

Two current theories of vowel normalization imply that vowel formant frequencies of one speaker can be related to those of another by a simple translation along a tonotopic frequency axis [T. M. Nearey, J. Acoust. Soc. Am. 85, 2088–2113 (1989)]. Theory A uses a Bark scale, while theory B uses a log-frequency (musical) scale. To compare these theories, a baseline continuum is first constructed that spans the range appropriate for the front vowels of an adult male speaker. In test series A, the formant frequencies are translated upward by 1.2 Bark, corresponding to the difference between average adult male and average female and child values from a vowel data base. In test series B, an upward translation of 0.2 natural log units is made, based on the corresponding average values in a log space. (Fundamental frequencies are also changed by the appropriate average value in the appropriate scale.) Each theory predicts its associated continuum will be identical to the baseline set, while that of the rival theory will show systematic distortion. Results of perceptual experiments will be presented that evaluate the phonetic distortion of the two test continua empirically.

Acoustic elements of speechlike stimuli are reflected in surface recorded responses over the guinea pig temporal lobe

Auditory evoked potentials measured from the guinea pig temporal lobe surface reflect acoustic elements of synthesized speech syllables. Eliciting stimuli included a four formant anchor stimulus /ba/, with a 40-ms formant transition duration. The other stimuli differed from /ba/ along simple acoustic dimensions. The /pa/ stimuli differed on a VOT continuum; /da/ stimuli had a higher frequency F2 onset; /wa/ had a longer (80 ms) formant transition duration; and /bi/ differed in three vowel formant frequencies. The /ba/ and /da/ onset response latencies decreased systematically with increasing F2 onset frequency. The response to the /pa/ voicing increased in latency with increasing VOT and showed a physiologic discontinuity at VOT of 15-20 ms. Responses to /ba/ and /wa/ showed similar onset morphology but significant amplitude differences at latencies corresponding to vowel onset. Significant amplitude differences in /ba/ and /bi/ responses corresponded in latency to both consonant and vowel portions of the syllables. Similar to previous reports in the awake monkey for VOT, these results demonstrate in the anesthetized guinea pig that acoustic elements essential to speech perception are reflected in aggregate response of ensembles of cortical neurons.

Acoustic characteristics of speech in French children using multichannel cochlear implants.

Acoustic characteristics of speech production produced with and without auditory feedback from a multichannel cochlear implant was examined in eleven French speaking children with profound hearing losses. Subjects produced five repetitions of stimuli designed to contrast voice‐onset times and vowel formant frequencies. Samples were collected with the implant turned on and after a 10‐min period of the implant turned off. Stimuli were audio recorded, low‐pass filtered, and digitized at a 10‐kHz rate. VOT’s and formant frequencies were measured using Cspeech software. VOTs for the bilabial cognates, /b/ and /p/, revealed considerable overlap in temporal values when no auditory feedback was available. During conditions providing auditory feedback, VOTs shifted to more nearly normal values with minimal overlap. Vowel formant frequencies also shifted as a function of implant status; however, the patterns of shift differed across subjects and across feedback conditions. Data suggest pediatric users of cochlear i...

Trade-off between quantity and stress in Estonian folksong performance?

This paper is concerned with the issue of how speech and music prosodies are matched to each other in folksong performance. We consider speech and music to be two more or less autonomous structures which both are required to be accommodated in singing performance according to a certain modus vivendi. Under some conditions, this necessity for coexistence may lead to a conflict between the two media which, as we believe, is caused by the different nature in which speech and music employ the acoustical continua of pitch, duration, and timbre. In speech, contrastive sounds phonemes - are primarily built up by exploiting the differences in sound spectrum (for example, vowels can be distinguished by differences in their lower formant frequencies). In a melody, however, the pitch dimension is of primary importance for its structure. The building-blocks for melodies are the scale steps which result from discretization of the frequency continuum into a number of separate levels. It seems reasonable to suppose that a desirable result of the coexistence of speech and music prosodies in singing would be as perfect match as possible between the two prosodies, unless certain existential« requirements from the side of one medium prevent reaching this. A number of studies about the relationship between the two prosodies in singing may be found in the literature which seem to justify this prediction. In tone languages like Chinese or Japanese, linguistically relevant tone patterns tend to be matched to melodic contours in music (Yung 1983). In languages like Indo-European where contrastive opposition exists between stressed and unstressed syllables, linguistic stress patterns tend to coincide with the metrical structure of stressed and unstressed patterns in music (Palmer & Kelly 1992). On the other hand, a classical example of lack of fit between the two prosodies is singing of high notes by female opera singers where it is physically impossible to keep the vowel formant frequencies at the vicinity of their etalon values from speech, because of the high fundamental frequency (Sundberg 1987). In this paper, we will address the issue of the goodness-of-fit between metrical and word stress in old Estonian folksong repertoire. Estonian is a Finno-Ugric language which belongs to a larger group of the Uralic languages. Stress in Estonian words always falls on the first syllable. An important characteristic of Estonian phonology is use of contrastive duration. There are three contrastive quantity degrees in standard Estonian words, and the difference between them is semantically relevant. The three degrees are known as short, long, and overlong. The difference between long and overlong degrees in most cases is not indicated in written language. In spoken language, differences between the three quantity degrees are manifested in speech by means of the ratio of the duration of the initial syllable to the duration of the second syllable in a word (Lehiste 1968). An important characteristic of the Estonian folksong repertoire is the relative independence of the text and the melody corpora. Almost every text from the former set may be combined with an arbitrary melody from the latter set. This implies that there must exist strong structural constraints which cause such an interplay between texts and melodies to be available to a performer. The necessary structural framework is provided by the metre which in Estonian folksongs is based on the contrast of long and short, rather than stressed and unstressed syllables. The standard so-called

Variability of vowel formant frequencies and speaker normalization: The cases of Mandarin and Taiwanese

Formant frequencies of vowels produced by male and female speakers of Mandarin and Taiwanese were measured. Twenty Mandarin-Taiwanese bilingual subjects produced three tokens of each vowel in the /lV/ (for Mandarin) and /kV/ (for Taiwanese) contexts. Like previous studies with other languages, women’s formants were higher than men’s in frequency, that high vowels had F1 values lower than low vowels, and that front vowels had F2 values higher than back vowels. Bark-difference measures were obtained from the Bark-transformed formant frequencies; that is F0 was subtracted from F1, F1 from F2, F2 from F3, F2 from F4, and F3 from F4. Two Bark-difference measures (i.e., F1−F0, F3−F2) correspond well to vowel height and point of articulation, respectively. Most importantly, high vowels have F1−F0 values within 3.5 Bark and front vowels have F3−F2 differences also within 3.5 Bark. These transformations largely reduced the between-speaker variability related to the gender factor. Results of these Mandarin and Taiwanese vowel data agree with English data [A. K. Syrdal and H. S. Gopal, J. Acoust. Soc. Am. 79, 1086–1100 (1986)] and support the universality of the category-independent auditory theory of normalization.

Vowel formant discrimination

This series of experiments examines the relation between the basic capabilities of the peripheral auditory system to process speech and the classification of speech at more central levels. Psychophysical techniques have been applied to determine listeners’ abilities to detect, discriminate, and identify speech sounds. The current series of experiments established that thresholds for the discrimination of vowel formants are constant in the F1 region with a value of F=14 Hz and in the F2 region increase linearly with increasing frequency. This presentation describes three directions of our recent research. First, the effects of various stimulus parameters, including fundamental frequency and consonantal context, on thresholds were explored. Second, excitation pattern models have been shown to account for the significant effects of vowel formant frequency and fundamental frequency on F thresholds. Third, an initial experiment compared formant discrimination to vowel identification for listeners with normal hearing and moderate hearing impairment. Results demonstrated that for the impaired listeners, reduced ability to identify vowels was partially predicted by elevated discrimination thresholds in the F2 region. [Work supported by NIDCD.]

Acoustic specification of North German vowels produced in citation and sentence contexts

As part of a larger study of the effects of speech style and consonantal context on the acoustic specification and cross-language perceptual similarity of vowels, this study examined differences in acoustic parameters of North German (NG) vowels in hVp syllables produced in citation-form (as lists) and in the carrier sentence ‘‘Ich habe—gesacht.’’ Two instances of each of 14 vowels produced by 4 male speakers (Kiel dialect) spoken in each context were analyzed. Vowel formant frequencies (measured at syllable midpoint) showed some undershoot of F1 in sentence context relative to citation context; F2 undershoot was minimal. Duration differences between spectrally similar tense-lax pairs varied as a function of vowel height in both citation and sentence contexts. Overall, tense-lax duration differences were reduced in sentence context relative to citation form. Individual differences across speakers in both F1/F2 targets and intrinsic durations were noted. Distances in F1/F2 space could not account for perceptual assimilation patterns of AE listeners [see S. Trent et al., J. Acoust. Soc. Am., this session]. These results have implications for conceptions of phonetic prototypes used to account for cross-language similarity. [Research supported by NIDCD.]

Formant frequencies of Hindi vowels in /hVḍ/ and C1VC2 contexts

Formant frequencies of Hindi vowels in /hVḍ/ and C1VC2 contexts for words spoken in isolation have been measured from sonagrams for two male and one female adult speakers and a spectrographic normalized coefficient of variance (C.V.) for F0 to F4 has been presented in a tabular form.

Acoustic characteristics of American English vowels

This study was designed as a replication and extension of the classic study of vowel acoustics by Peterson and Barney (PB) [J. Acoust. Soc. Am. 24, 175–184 (1952)]. Recordings were made of 50 men, 50 women, and 50 children producing the vowels /i, i, eh, æ, hooked backward eh, inverted vee), a, open oh, u, u/ in h–V–d syllables. Formant contours for F1–F4 were measured from LPC spectra using a custom interactive editing tool. For comparison with the PB data, formant patterns were sampled at a time that was judged by visual inspection to be maximally steady. Preliminary analysis shows numerous differences between the present data and those of PB, both in terms of average formant frequencies for vowels, and the degree of overlap among adjacent vowels. As with the original study, listening tests showed that the signals were nearly always identified as the vowel intended by the talker.

Trans-syllabic spread of consonantal effects on vowel formant frequencies

Formant-based synthetic speech is less robust in noise than natural speech and is often criticized as ‘‘robot-like.’’ One reason may be the failure to model nonlocal spectral variation induced by segments in other syllables. This study investigates how a single consonant can affect vowel formant frequencies in nonadjacent as well as adjacent syllables. Sequences of /schwa u i schwa/ with intervening consonants were embedded in carrier phrases to give quasimeaningful sentences. The medial consonant was /z/ or /r/. Flanking consonants were either all /b/, or /d/ in the syllables before and after the medial consonant (/bschwabu{zr}ibschwab/, /bschwadu{zr}idschwab/). Primary stress was on either the second syllable, or the first and third. Steady-state or mid-point frequencies of F1–F3 were measured from LPC spectra. Preliminary results confirm expectations: /r/ engenders lower formant frequencies than /z/ in less-stressed, all-b contexts, even in nonadjacent syllables. Vowel stress and /d/ contexts tend to block the spread of differences due to /z/ or /r/, presumably because the tongue is more constrained. Many of these differences are audible; their contribution to robustness and naturalness of synthetic speech will be described. [Work partly supported by Telia Promotor Infovox AB.]

Pilot study on cross-language transformation of vowels

This pilot study explores an aspect of speaker individuality, the individual’s vowel space, when one person speaks two languages. It is asked whether a speaker’s characteristic range of vowel formant frequencies can be measured in one language and used, in conjunction with vowel data found in the literature, to model the speaker’s vowel space in the other language. One practical application of the results would be to make a computer speech recognition system used as a pronunciation trainer speaker-adaptive. Vowel data are being gathered from a small set of bilingual speakers of English and Italian and English and Mandarin Chinese. Linear and nonlinear transformations for converting the F2 vs F1 plots of each of the bilinguals between the appropriate languages will be explored, in an approach similar to those of Disner [J. Acoust. Soc. Am. 67, 253–261 (1980)] and DiBenedetto [ICSLP 92, 579–582 (1990)]. The adequacy of the transformations and of the simple vowel representation employed, i.e., F1 and F2 midpoints, will be discussed.

デジタル音声処理による上顎欠損症例の音声分析

Vowel sounds spoken by patients having maxillary defects were compared and evaluated with a digital signal processor that was capable of automatically recognizing the formant frequencies of Japanese vowels before and after wearing a maxillary prosthesis.The automatic recognition rate for formant frequencies for this system was 91.6%, and immediate recordings and analyses of sound waveforms were recorded on floppy discs.Results showed that a sexual difference of formant frequency existed among normal subjects, but no age difference for those over 20 years of age was shown. Patients having maxillary defects showed the vowel sounds for /i/, /u/, and /e/ to be located mostly outside the normal equiprobability ellipsoid. The F 1 value for patients having maxillary defects tended to increase slightly for /i/ and increase as well as decrease for /e/. Vowel sounds /a/, /u/, and /o/ remained within the normal range. On the otherhand, the F 2 value tended to increase for /a/, /u/ and /o/, and decrease for /i/and /e/ for patients having maxillary defects.In conclusion, compared to H 1 type maxillary defects, H 3 and H 4 type showed most vowel sounds to be located outside the normal equiprobability ellipsoid. However, the vowel sounds were found to be within the normal range when the patients wore their dentures.

The effects of auditory feedback from the nucleus cochlear implant on the vowel formant frequencies produced by children and adults.

Cochlear implants provide an auditory signal with which profoundly deaf users may monitor their own speech production. The vowel production of two adults and three children who used the Nucleus multiple-electrode cochlear implant was examined to assess the effect of altered auditory feedback. Productions of words were recorded under conditions where the talkers received auditory feedback (speech processor turned on) and where no auditory feedback was provided (speech processor turned off). Data were collected over 3 days at weekly intervals. First and second formant frequencies were measured and the data were analysed to assess significant differences between auditory feedback conditions, vowel context, and data collection points. Overall, the results varied across talkers, across the data collection days, and depended on the consonant environment of the vowel. However, two effects of auditory feedback were noted. First, there was a generalized shift in first formant frequencies between the processor on and processor off conditions across three of the five subjects, but the shift differed in direction for each subject. Second, for three of the five talkers, the two front vowels /epsilon/ and /i/ were more neutralised in the absence of auditory feedback. However, this effect was less pronounced than that noted by previous studies.

Effects of speaking rate, focal stress, and sentence position on spectral characteristics and vowel duration

Speaking rate, focal stress, and sentence position can affect the duration and spectral characteristics of vowels [e.g., B. Lindblom, J. Acoust. Soc. Am. 35, 1773–1781 (1963)]. However prior research has been less than conclusive in characterizing effects of these factors [cf. e.g., T. Gay, J. Acoust. Soc. Am. 63, 223–230 (1978)]. Previous findings [D. M. Behne and L. C. Nygaard, J. Acoust. Soc. Am. 90, 2254(A) (1991)] have confirmed the combined effects of these factors on vowel duration and have shown that vowels can be independently affected by speaking rate, focal stress, and sentence position. The goal of this paper is to characterize the effects of these factors on formant frequencies of vowels and relate them to effects on vowel duration. Using the vowels /i, o, a, ae/, conversations were developed with /kVt/ and /kVd/ as target words. In each conversation a target word occurred in initial and in final sentence position and was either focused or nonfocused by the discourse. Twelve subjects produced each conversation at three speaking rates. The present results demonstrated no effect of either speaking rate or focal stress on the spectral characteristics of the vowels, but relatively high first formant frequency and low second formant frequencies for vowels in final sentence position compared to initial sentence position. These findings verify that speaking rate, focal stress, and sentence position can concurrently influence syllable‐internal timing with distinct acoustic effects. The results will be discussed in terms of vowel reduction and undershoot.

Modeling format frequency discrimination

Changes in the formant frequencies of steady-state vowels can produce level changes in harmonic components throughout the bandwidth of the stimulus. To successfully model thresholds of formant frequency discrimination, however, it is necessary to determine whether level changes in components spectrally remote from the altered formant affect discrimination performance. Therefore, the present experiments investigate which harmonics affect performance in order to evaluate excitation-pattern models of formant frequency discrimination. Synthetic tokens of the vowels /u/ and /e/ were generated by digitally summing the first 24 harmonics of a 200-Hz fundamental. Minimal uncertainty thresholds were obtained for F1 increments in /e/ and F2 decrements in /u/ when: (1) all harmonics were allowed to vary; (2) harmonic level variations were restricted to a subset of components spectrally close to the altered formant; and (3) harmonics remote from the shifted formant were excluded from the stimulus. Results indicated that thresholds for conditions (1) and (2) did not differ until level variations were restricted to between two and three components. Thresholds for condition (3) were not significantly different than those obtained for vowels with a full complement of harmonics. Excitation-pattern models of the results will be discussed. [Research supported by NIH.]

Masking of tones and formant frequencies

The purpose of this study was to find out whether results of auditory masking theory, established primarily for pure tones, can be used to predict masking of vowel formant frequencies by broadband noise maskers. For this purpose, three sets of 2AFC experiments were conducted. In each experiment, four frozen white-noise waveforms (with similar statistical properties) were used. The task in the first set of experiments, was tone-in-noise detection in which the tones were located at 1100 and 1800 Hz. In the second set, one-formant synthetic vowels were used in vowel-in-noise detection tasks. The formant was located at 1100 Hz in one experiment, and at 1800 Hz, in another. Finally, listeners were presented with pairs of four-formant synthetic vowels in noise: one pair had identical vowels and the second vowel in the second pair had a missing second formant frequency (F2). Subjects were instructed to choose the interval thought to contain the different pair of vowels. The four-formant synthetic vowel was either /ɑ /-like (formants at 625, 1100, 2700, and 3250 Hz) or /schwa/-like (formants at 500, 1800, 2700, and 3500 Hz). It was found that, on the average, the masked thresholds of tones and of formant frequencies were predicted well from the signal-to-noise ratio in a critical band centered around the tone or the formant frequency. [Work supported by NIH and NSF.]

Formant frequencies in Estonian folk singing.

Formant frequencies in an old Estonian folk song performed by two female voices were estimated for two back vowels /a/ and /u/, and for two front vowels /e/ and /i/. Comparison of these estimates with formant frequencies in spoken Estonian vowels indicates a trend of the vowels to be clustered into two sets of front and back ones in the F1/F2 plane. Similar clustering has previously been shown to occur in opera and choir singing, especially with increasing fundamental frequency. The clustering in the present song, however, may also be due to a tendency for a mid vowel to be realized as a higher-beginning diphthong, which is characteristic of the North-Estonian coastal dialect area where the singers come from. No evidence of a "singer's formant" was found.

Effects of distorted auditory feedback on speech.

A research program for acoustically measuring changes in speech production will be described. Here, F0 and spectrographic measurements target acoustic correlates of consonant, vowel, and syllabic stress features in 45 phonemically balanced words. To date, 5 Nucleus-22 multichannel cochlear implant patients have participated (at least) prestimulation, immediately after stimulation, 3–6 months after stimulation, and 1-year after stimulation. Commonalities among patients include: a rapid increase and stabilization of F0 (in females) to normal values; immediate and continuing changes in vowel and syllable durations and in relative amounts of breathy noise and low-frequency energy, gradual changes in vowel formant frequencies with movement still observed at 1 year; and changes in consonant properties (such as F1 cutback for voiceless stops) emerging for the first time at 1 year. Results relate to (1) the potential role of critical periods in the effect of auditory feedback on speech habits, (2) the relative susceptibility of speech features to modification, and (3) the likely auditory parameters available from Nucleus-22 WSP III stimulation. [Work supported by NIDCD.]